Ripsnort, the answer to your question is that it depends. If administered in a scientific manner, polls can be very, VERY reliable. If administered with a particular motivation in mind, obviously the results must be viewed with some scrutiny.
Several things make a poll scientific. First, the sample size matters. There is a tradeoff between sample size and margin of error. The smaller the sample size, the larger the margin of error (and the lower the costs of administering the survey). The relationship isn't linear, however, and you'll experience diminishing returns from increased sample sizes. Ideally, a survey should have between 1500 to 2000 respondents for a margin of error at or below plus/minus 2%. This is Newsweek's biggest problem, as they tend to survey about 500 to 700 people for their polls.
Second, a survey must be randomly administered, or as near-randomly administered as possible. The problem here is that most polling firms won't tell you how they drew up their sample. Ideally you'd want a completely random sample; as this is impossible, a design such as random-sample clustering works well. If I'm not mistaken, I believe this method breaks down the country into various regions, and these regions into various areas, and these areas into smaller areas, on and on down to the individual. Randomization then occurs at each stage.
Finally, polls must have valid questions. A poll asking, "What are your views on abortion? Are you in favor of it, against it, undecided, or none of these?" is far more scientific than a poll asking, "Do you support the mass execution of unborn human lives through the atrocity of abortion?" Be wary of polls that don't disclose the exact wording of the questions asked to interviewees. As a matter of priming, question order also may influence answers. Unfortunately, it's nearly impossible for the casual reader to discern the order in which questions were asked.
That said, Gallup is a decent organization, and certainly the best known among polling firms. As it is under so much scrutiny to get things right, chances are that their polls are conducted with as much professionalism as possible.
Probably the best survey design and administration you'll find, period, is the annual National Election Studies(NES)out of the University of Michigan. The link there, if you're curious, is
http://www.umich.edu/~nes/ Hope this helps,
Todd/DMF
